Recurrent hotspot mutations in HRAS Q61 and PI3K-AKT pathway genes as drivers of breast adenomyoepitheliomas

1Adenomyoepithelioma of the breast is a rare biphasic tumor composed of epithelial and myoepithelial cells, which typically displays a benign clinical course, but may recur locally and/or metastasize. Phenotypically, adenomyoepitheliomas are heterogeneous. The epithelial component may express estrogen receptor (ER) and progesterone receptor (PR); however, a subset of adenomyoepitheliomas lacks the expression of hormone receptors altogether. Both the epithelial and myoepithelial compartments can expand and undergo malignant transformation, histologically characterized by nuclear atypia, mitotic activity, and/or necrosis,,. Importantly, however, metastases have been documented even in cases lacking a histologically overt malignant component. Interestingly, most invasive breast cancers arising in adenomyoepitheliomas display a triple-negative phenotype (ER-, PR- and HER2-negative) and metaplastic features.

2The genomic landscape of breast cancers has been extensively investigated (reviewed in Ng et al.). Large-scale massively parallel sequencing studies have revealed that breast cancers display a complex repertoire of somatic mutations, that TP53 (37%), PIK3CA (36%), and GATA3 (11%) are the only three genes recurrently mutated in >10% of unselected breast cancers, and that the repertoire of somatic mutations differs between ER-positive and ER-negative disease; however, no pathognomonic mutations underpinning ER-positive or ER-negative breast cancers have been identified,. These analyses, however, primarily focused on the common forms of breast cancer–, whereas the genetic characteristics of rare forms of breast tumors, including those with myoepithelial differentiation, remain largely unexplored. In fact, genomic analyses of adenomyoepitheliomas, based on case reports or small series of cases, have demonstrated the presence of a t(8;16)(p23;q21) chromosomal translocation in a single case or a TP53 R270C missense mutation in another. The landscape of somatic genetic alterations of adenomyoepitheliomas, however, has yet to be characterized.

3Here we report on a combination of whole-exome (WES) and targeted capture massively parallel sequencing analyses that revealed that adenomyoepitheliomas are genetically heterogeneous and that, akin to common-type breast cancers, their repertoire of somatic mutations vary according to their ER status. In ER-negative adenomyoepitheliomas, recurrent HRAS Q61 hotspot mutations co-occur with mutations affecting PI3K pathway genes. In non-malignant triple-negative breast epithelial cells with or without a somatic knock-in of a PIK3CA H1047R mutation, forced expression of mutant HRAS promotes growth advantage, the acquisition of features consistent with myoepithelial differentiation, and activation of PI3K-AKT and MAPK signaling pathways, likely acting as a driver of ER-negative adenomyoepitheliomas.

4Adenomyoepitheliomas were retrieved from the authors’ institutions after approval by the local Institutional Review Boards (IRBs). Upon central histologic review, 43 cases were considered bona fide adenomyoepitheliomas, of which 18 (42%) displayed atypical histologic features suggestive of a more aggressive behavior (i.e. marked nuclear pleomorphism, high mitotic rate, and/or necrosis; Fig. 1, Supplementary Data 1). Immunohistochemical analysis revealed that 16 adenomyoepitheliomas (37%) lacked ER expression (Fig. 1g, Supplementary Figs. 1a−l), a feature that was significantly associated with nuclear pleomorphism, increased mitotic rate, and higher Ki-67 labeling indices (P < 0.05, Fisher’s exact tests, Fig. 1h−i, Supplementary Data 1, Supplementary Fig. 1). AR was expressed in ≥10% of tumor cells (the cut-off employed to select patients for potential anti-androgen therapy) in all but one (26/27, 96%) ER-positive adenomyoepitheliomas, but in 9/16 (56%) ER-negative adenomyoepitheliomas (P < 0.01, Fisher’s exact test, Fig. 1i, Supplementary Figs. 1c, i). All adenomyoepitheliomas studied lacked HER2 protein overexpression and displayed focal p53 expression suggestive of an unaltered TP53 status (Fig. 1i, Supplementary Data 1, Supplementary Figs. 1f, l). As expected, the vast majority of adenomyeopitheliomas (88%) displayed strong p63 expression in the myoepithelial component (Supplementary Figs. 1d, j).

5Seven adenomyoepitheliomas (16%) were associated with invasive carcinoma: six present in the primary tumor and one in the ipsilateral breast recurrence (Fig. 1e, Supplementary Data 1, Supplementary Figs. 1m−x). The ER status of paired adenomyoepitheliomas and carcinomas was concordant in all but one case; in the ER-positive AM46, the invasive carcinoma of spindle cell metaplastic type lacked ER expression. Three ER-negative adenomyoepitheliomas developed local recurrences and/or had axillary lymph-node metastases (Supplementary Fig. 1, Supplementary Data 1). Consistent with previous reports of biphasic metastases of adenomyoepitheliomas, the metastases observed in these cases retained the epithelial−myoepithelial phenotype (Fig. 1, Supplementary Fig. 1f), suggesting that at least in some cases, the epithelial and myoepithelial cell populations likely share a common cell of origin with dual-lineage potential.

6To define the genomic landscape of adenomyoepitheliomas, DNA samples extracted from 31 tumor-normal pairs of adenomyoepitheliomas (Supplementary Data 1) were subjected to WES (n = 10, median depth of coverage of tumor 159× (range 117×–167×) and normal 95× (range 69×–175×) samples) or targeted capture massively parallel sequencing due to limited yields of DNA available (n = 21, median depth of coverage of tumor 434× (range 252×–749×) and normal 401× (range 83×–897×) samples), using the Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT). This sequencing assay targets all coding regions of 410 key cancer genes and intronic and regulatory regions of selected genes (Supplementary Data 2). In addition, the TERT gene promoter region was investigated in all cases by MSK-IMPACT and/or Sanger sequencing.

7Adenomyoepitheliomas displayed a low mutation burden, with a median of 17 (range 5−63) somatic mutations as defined by WES, of which 13.5 (range 4–47) were non-synonymous, whereas the MSK-IMPACT assay detected a median of 2 (range 0–7) somatic mutations, of which 2 (range 0–6) were non-synonymous (Supplementary Data 2). Compared to common forms of invasive breast cancers from The Cancer Genome Atlas (TCGA) dataset (median 46, range 10−396), adenomyoepitheliomas subjected to WES were found to have significantly fewer somatic mutations (P < 0.0001, Wilcoxon test). By WES, ER-negative adenomyoepitheliomas displayed a numerically but not statistically significant higher number of somatic mutations than ER-positive adenomyoepitheliomas (22 (range 6–36) versus 14 (range 5–63), P > 0.1, Wilcoxon test). In agreement with the immunohistochemistry analysis, TP53, the gene most frequently mutated in breast cancer and mutated in up to 80% of ER-negative breast cancers, was not found to be altered in any of the adenomyoepitheliomas analyzed in this study.

8Despite the low mutation burden, mutational analysis revealed recurrently mutated genes, including PIK3CA (16/31, 52%), HRAS (8/31, 26%), AKT1 (4/31, 13%), and PIK3R1 (4/31, 13%). The PIK3CA mutations preferentially affected mutation hotspots (six H1047R, five E545K, and one E542K hotspot mutations). All HRAS mutations affected the mutation hotspot Q61, and all AKT1 mutations were the E17K hotspot mutation (Fig. 2, Supplementary Data 3). In four adenomyoepitheliomas (AM1, AM27, AM43, and AM46), dual PIK3CA mutations were identified, and in AM5 and AM52, dual PIK3R1 small deletions were detected. Of the 38 mutations affecting these four genes detected by WES or MSK-IMPACT, 34 (89%) were found to be clonal by ABSOLUTE (Supplementary Fig. 2a, Supplementary Data 3). Additional cancer genes recurrently mutated in adenomyoepitheliomas included TERT (4/31, 13%, all hotspot promoter mutations) and PRKAR1A (2/31, 6%, Fig. 2, Supplementary Data 3). Likely pathogenic mutations affecting single tumors included an ERBB3 D297Y hotspot mutation in a PIK3CA-mutant and HRAS-wild-type adenomyoepithelioma (Fig. 2, Supplementary Data 3).

9Importantly, however, differences were observed in the repertoire of somatic mutations of adenomyoepitheliomas: AKT1 hotspot mutations were solely found in ER-positive lesions (24% versus 0%, P = 0.1074, Fisher’s exact test), PIK3R1 small deletions were only detected in ER-negative adenomyoepitheliomas (29% versus 0%, P = 0.0318, Fisher’s exact test), and HRAS Q61 hotspot mutations were restricted to ER-negative adenomyoepitheliomas (57% versus 0%, P = 0.0004, Fisher’s exact test) and always co-occurred with PIK3CA or PIK3R1 somatic mutations (Fig. 2, Supplementary Data 3).

10To validate the HRAS, PIK3CA, and AKT1 mutations identified by WES or MSK-IMPACT, Sanger sequencing of the hotspot loci was performed in the initial 31 cases and 12 additional adenomyoepitheliomas, for which sufficient DNA could not be obtained for massively parallel sequencing. All PIK3CA, HRAS, and AKT1 hotspot mutations were validated in the index cases, and four, one, and one mutations were detected in PIK3CA, HRAS, and AKT1, respectively, in the 12 additional cases (Fig. 3a, Supplementary Fig. 2b, Supplementary Data 3). Mutations affecting PIK3CA, AKT1, and PIK3R1 were found to be significantly mutually exclusive (n = 43, P = 0.0004, CoMEt exact test), whereas all HRAS mutations significantly co-occurred with PIK3CA or PIK3R1 mutations (n = 43, P value for co-occurrence = 0.025, Z test).

11In ER-positive adenomyoepitheliomas, mutations affecting PIK3CA or AKT1 were found in 59% (16/27) of cases, whereas in ER-negative adenomyoepitheliomas, mutations affecting PIK3CA or PIK3R1 were detected in 81% of the cases tested (Fig. 3a). Within ER-positive adenomyoepitheliomas, the presence of PIK3CA or AKT1 mutations was significantly associated with marked nuclear pleomorphism and moderate-to-high mitotic rate (both 46% versus 0%, P = 0.0216, Fisher’s exact tests), whereas among ER-negative adenomyoepitheliomas, HRAS mutations were significantly associated with necrosis (67% versus 0%, P = 0.0114, Fisher’s exact test) and high mitotic rate (56% versus 0%, P = 0.0337, Fisher’s exact test).

12In the whole dataset (n = 43, Fig. 3), all HRAS (Q61R/K), all AKT1 (E17K), and 16/24 PIK3CA mutations detected were known activating hotspot mutations (Fig. 3a). Of the eight non-hotspot PIK3CA mutations, six affected hotspot residues and all were predicted to be likely pathogenic (Supplementary Data 3). While PIK3CA mutations are common in breast cancer (approximately 35%), mutations in AKT1 and HRAS were found to be significantly less frequent in unselected breast cancers than in the adenomyoepitheliomas studied here (P < 0.05, Fisher’s exact tests, Fig. 3b, Supplementary Data 4),. In fact, HRAS mutations were not found in any of the common-type breast cancers included in the TCGA study (n = 507), and detected only in one of 560 (0.2%) breast cancers from the whole-genome sequencing analysis carried out by the International Cancer Genome Consortium (ICGC).

13Taken together, adenomyoepitheliomas of the breast are characterized by low mutation rates and the lack of TP53 mutations. Mutations affecting PI3K-AKT pathway-related genes are frequent across ER-positive and ER-negative adenomyoepitheliomas. HRAS Q61 hotspot mutations, which are remarkably rare in breast cancers, are restricted to ER-negative adenomyoepitheliomas and associated with atypical histology indicative of a more aggressive phenotype.

14Genome-wide copy number analysis revealed a diploid/near-diploid genome in the 31 adenomyoepitheliomas analyzed here (Supplementary Fig. 3a), which harbored fewer copy number alterations (CNAs) than common forms of breast cancers from TCGA (Supplementary Fig. 3b). Among the most frequent CNAs were losses of 6p22 (6/31, 19%), 9p21 (CDKN2A, 4/31, 13%), and 4q31 (INPP4B, 2/31, 6%) and gains of 12p12.3 (ETV6, 5/31, 16%) and 5p15 (TERT, 4/31, 13%, Supplementary Fig. 3a). Losses of 9p21, which have been previously linked to an unfavorable phenotype in breast cancer, were only detected in adenomyoepitheliomas lacking ER expression (29% versus 0%, P = 0.0318, Fisher’s exact test, Supplementary Fig. 3c), and displaying histologic features associated with an unfavorable clinical behavior (i.e., necrosis and/or high mitotic rate; 36% versus 0%, P = 0.0105, Fisher’s exact test). An exploratory, hypothesis-generating genome-wide analysis revealed that ER-negative adenomyoepitheliomas displayed a significantly higher number of CNAs than ER-positive lesions (P = 0.0374, Wilcoxon test, Supplementary Fig. 3d), and adenomyoepitheliomas with atypical histologic features suggestive of a more aggressive behavior harbored significantly more CNAs than those without (P = 0.0051, Wilcoxon test, Supplementary Fig. 3e). Moreover, gains of chromosomes 7 and 8 were restricted to ER-negative adenomyoepitheliomas; however, these were low level gains. Further studies are warranted to define their biological and clinical significance.

15Homozygous deletions and high-level amplifications were rarely found in adenomyoepitheliomas. Of note, recurrent homozygous deletions affecting CDKN2A were found in 3/14 (21%) ER-negative adenomyoepitheliomas (Fig. 2, Supplementary Fig. 3f) and, interestingly, these three cases were found to progress to carcinoma (100% versus 9%, P = 0.011, Fisher’s exact test). Amplifications of oncogenes (e.g., FGFR2 and PIK3CA) were found in individual cases (Fig. 2, Supplementary Fig. 3f), but none of the adenomyoepitheliomas studied harbored HER2 gene amplification, consistent with the lack of HER2 protein overexpression by immunohistochemistry (Fig. 1i).

16To assess whether progression to a malignant phenotype is associated with the acquisition of additional somatic genetic alterations and/or clonal selection, we analyzed separately microdissected components of the primary tumor, locally recurrent tumor, and/or lymph node metastases of three patients with adenomyoepitheliomas (Supplementary Figs. 1m−x).

17The separately microdissected components of AM8 and AM46 were subjected to MSK-IMPACT sequencing. In AM8, this analysis revealed clonal HRAS, PIK3CA, and TERT promoter mutations and CDKN2A homozygous deletions in all components analyzed, consistent with these being truncal (i.e. present as clonal alterations in all components analyzed) genetic events (Fig. 4a, Supplementary Data 3). In AM46, two clonal PIK3CA mutations (Fig. 4b, Supplementary Data 3) were found to be truncal. The adenomyoepithelioma component likely displayed intra-tumor genetic heterogeneity, given that it harbored a subclonal EPHB1 missense mutation and a subclonal TERT promoter hotspot mutation. Interestingly, the latter was found to be clonal in the carcinoma component of this case.

18These analyses suggested the presence of intra-tumor genetic heterogeneity within adenomyoepitheliomas and potential clonal selection in the progression to carcinoma. To investigate these hypotheses, we performed WES analysis of the separately microdissected components of AM5. Based on the analysis of the validated somatic genetic alterations, the adenomyoepithelioma and carcinoma components of the primary tumor and the axillary lymph node metastasis shared 11 clonal truncal mutations, including an HRAS and two PIK3R1 mutations, and a CDKN2A homozygous deletion (Fig. 4c, Supplementary Data 3), and truncal CNAs including losses of 9p and 9q. Additional private clonal mutations were found in each of the components, including a frameshift mutation affecting NOTCH1, which was coupled with loss of heterozygosity of its wild-type allele due to the truncal 9q loss, and a loss of chromosome 10 restricted to the carcinoma. These findings are consistent with the presence of intra-tumor genetic heterogeneity within the components analyzed, and suggest that each component may have undergone branching evolution (Fig. 4c). Importantly, however, the biologic significance of the private genetic alterations identified remains to be defined.

19Taken together, our multi-region genetic analyses demonstrate that adenomyoepitheliomas and their respective carcinomatous or metastatic components display remarkable similarities in regards to their mutational and CNA profiles, with a limited number of known driver genetic alterations enriched or solely detected in the carcinomatous or metastatic components of each case. In addition, this analysis is consistent with the notion that HRAS Q61 hotspot mutations, PIK3CA mutations, PIK3R1 mutations, and CDKN2A homozygous deletions, when present, likely constitute truncal genetic events, whereas TERT promoter mutations may constitute early or late events in the development and/or progression of adenomyoepitheliomas.

20Given the high prevalence of HRAS Q61 and PIK3CA mutations in ER-negative adenomyoepitheliomas, and that these mutations co-occur, we sought to define the functional impact of these mutations in non-malignant ER-negative breast epithelial cells. Given that adenomyoepithelioma cell lines or patient-derived xenografts are not commercially available, we investigated the functional impact of forced expression of the HRAS Q61R mutation using the non-malignant breast epithelial cells MCF-10A and MCF-12A. We reasoned that these cell lines would constitute an adequate model for ER-negative adenomyoepitheliomas given their triple-negative phenotype, the fact that they are TP53 wild-type, and MCF-10A cells harbor a CDKN2A homozygous deletion, which was the fourth most frequent somatic genetic alteration identified in ER-negative adenomyoepitheliomas (Fig. 2b). In addition to the parental MCF-10A cells (MCF-10AP), to define the impact of concurrent HRASQ61R and PIK3CA hotspot mutations (i.e., PIK3CAH1047R or PIK3CAE545K), we employed MCF-10A cells where the oncogenic PIK3CAH1047R or PIK3CAE545K were knocked-in (MCF-10AH1047R and MCF-10AE545K, respectively).

21As expected, forced expression of mutant HRASQ61R in monolayer MCF-10AP, MCF-10AH1047R, and MCF-12A cells resulted in upregulation of GTP-bound HRAS (Supplementary Fig. 4a) and increased ERK1/2 phosphorylation (T202/Y204) as compared to the empty vector (EV) control (Supplementary Fig. 4b). We next investigated the oncogenic impact of HRAS Q61R on non-malignant breast epithelial cells. Forced expression of mutant HRASQ61R in MCF-10AP, MCF-10AH1047R, and MCF-12A cells resulted in an increase in colony formation and cell proliferation as compared to EV or HRASWT in soft agar (Fig. 5a) and MTT assays (Fig. 5b), respectively. Moreover, forced expression of mutant HRASQ61R resulted in a significantly increased migration as compared to EV or HRASWT in the three cell lines, as demonstrated by a wound-healing assay (Fig. 5c, P < 0.05, unpaired t-tests). Consistent with the observations made using PIK3CAH1047R cells, forced expression of mutant HRAS in MCF-10AE545K cells also resulted in increased proliferation and migration, as compared to EV and HRASWT (Supplementary Fig. 5a−c). In addition, mammosphere formation assays revealed that forced expression of mutant HRASQ61R resulted in a significant increase in the number of spheres than forced expression of EV or HRASWT in all cell lines tested (Supplementary Fig. 4c, P < 0.05, unpaired t-tests).

22Functional experiments provided evidence that the HRAS Q61R mutation with or without a PIK3CA H1047R or E545K mutation results in increased proliferation and migration in non-malignant breast epithelial cells. Given that HRAS Q61 hotspot mutations are vanishingly rare in common forms of breast cancers (Fig. 3b), but present in the majority of ER-negative adenomyoepitheliomas and always in conjunction with PIK3CA or PIK3R1 somatic mutations, we posited that HRAS Q61 hotspot mutations would not only constitute an oncogenic driver of ER-negative adenomyoepitheliomas, but also play a role in the acquisition of an adenomyoepithelial phenotype. To determine the impact of HRAS and PIK3CA hotspot mutations on the differentiation of non-malignant breast epithelial cells, we assessed the expression levels of proteins preferentially expressed in epithelial or basal/myoepithelial cells of the breast, in MCF-10AP, MCF-10AH1047R, MCF-10AE545K, and MCF-12A cells grown in monolayer, and the phenotypic changes induced by HRAS and PIK3CA hotspot mutations when MCF-10AP, MCF-10AH1047R, and MCF-12A cells were grown in three-dimensional model systems (Fig. 6).

23In monolayer cultures, forced expression of mutant HRASQ61R in MCF-10AH1047R, MCF-10AE545K, and MCF-12A resulted in downregulation of E-cadherin, which is not expressed in normal myoepithelial cells, and upregulation of alpha-smooth muscle actin (αSMA), an integral component of the smooth muscle apparatus of myoepithelial cells and expressed in adenomyoepitheliomas, TIMP1, another marker of myoepithelial differentiation, and vimentin, which is also expressed in myoepithelial cells of normal breast and adenomyoepitheliomas, as compared to EV or forced expression of HRASWT (Fig. 6a, Supplementary Fig. 5d). In addition, forced expression of mutant HRASQ61R led to a significantly increased ∆N-p63 (p40)/TA-p63 ratio as compared to EV or forced expression of HRASWT (Fig. 6a, Supplementary Fig. 5d, P < 0.05, unpaired t-tests). It should be noted that ∆N-p63 is the p63 isoform preferentially expressed in cells with myoepithelial differentiation,, whereas TA-p63 is constitutively expressed at baseline in MCF-10A cells and reported to have anti-suppressive properties. Consistent with the notion that this phenomenon is related to the acquisition of a partial myoepithelial phenotype rather than epithelial-to-mesenchymal transition, the levels of cytokeratin 5 (CK5), which is expressed in epithelial and myoepithelial cells of normal breast and adenomyoepitheliomas, did not change accordingly (Fig. 6a, Supplementary Fig. 5d). Quantitative real-time reverse transcription PCR (qRT-PCR) confirmed the significantly higher levels of genes preferentially expressed in normal myoepithelial cells of the breast,, including ACTA2 (encoding αSMA), TIMP1, SPARC, and VIM (encoding vimentin), and significant downregulation of CDH1 (encoding E-cadherin) in MCF-10AP, MCF-10AH1047R, and MCF-12A cells expressing mutant HRASQ61R as compared to cells expressing HRASWT (Supplementary Fig. 4d, P < 0.05, unpaired t-tests). These observations suggest that forced expression of mutant HRASQ61R may be sufficient to induce the acquisition of at least a partial myoepithelial phenotype in non-malignant breast epithelial cells.

24Forced expression of mutant HRASQ61R in MCF-10AP, MCF-10AH1047R, and MCF-12A cells grown in three-dimensional organotypic cultures, resulted in a phenotype shift; from round, regular, polarized acinar structures with hollow lumina to irregular, multi-acinar structures connected through duct-like extensions with partially filled lumina (Fig. 6b), a phenotype reminiscent of myoepithelial colonies grown in Matrigel. Akin to human adenomyoepitheliomas that display markedly decreased E-cadherin expression in the myoepithelial cells as compared to the epithelial cells and express vimentin in the myoepithelial cell layer (Fig. 6c), immunofluorescence analysis of the acinar structures of MCF-10AP, MCF-10AH1047R, and MCF-12A cells demonstrated that forced expression of wild-type and mutant HRAS resulted in downregulation of E-cadherin and upregulation of vimentin (Fig. 6d). Importantly, however, in MCF10AH1047R and MCF12A cells, forced expression of mutant HRASQ61R resulted in vimentin expression in the abluminal layer, closely recapitulating its expression in the gland-like structures found in human adenomyoepitheliomas (Fig. 6c). In addition, forced expression of mutant HRASQ61R led to a consistent increase in the expression levels of calponin, another marker of myoepithelial differentiation, expressed in adenomyoepitheliomas, in all cell lines tested (Fig. 6e). qRT-PCR analysis of RNA extracted from acinar structures corroborated the downregulation of CDH1 and upregulation of SPARC and TIMP1 (Supplementary Fig. 4e).

25Taken together, these observations are consistent with the hypothesis that HRASQ61R may be sufficient for the acquisition of a partial myoepithelial phenotype in ER-negative non-malignant breast epithelial cells, and that this phenotype becomes more overt in the presence of a PIK3CA H1047R hotspot mutation.

26Co-occurrence of genetic alterations that induce activation of MAPK and PI3K-AKT pathways has been shown to result in stronger oncogenic potential than genetic alterations affecting either pathway alone. Given that cross-talks between both pathways occur and RAS is a positive regulator of AKT,, we sought to investigate whether HRAS Q61 mutations are preferentially acting via the PI3K-AKT or MAPK signaling cascades, and the effects of AKT (MK2206) and MEK (GSK212) pharmacological inhibition in HRASQ61R-expressing MCF-10AP and MCF-10AH1047R cells.

27We first assessed the phosphorylation of the downstream targets of RAS and AKT at baseline and upon treatment with AKT and MEK inhibitors (AKTi and MEKi, respectively) at different time-points (Fig. 7a, Supplementary Fig. 6a). Forced expression of HRASQ61R as compared to EV resulted in higher phosphorylation of MAPK signaling pathway components, as well as of markers downstream of AKT and mTOR, not only in MCF-10AP, but also in MCF-10AH1047R cells (Fig. 7a, Supplementary Fig. 6a). These findings suggest that activation of PI3K-AKT-mTOR pathway is a key consequence of the HRAS Q61R mutation, and that both HRAS Q61 and PIK3CA mutations may cooperate for strong activation of the PI3K-AKT-mTOR pathway in ER-negative adenomyoepitheliomas.

28Next, we conducted dose−response studies to assess the dependency of RAS-induced proliferation on MAPK or AKT signaling (Supplementary Fig. 6b). The median inhibitory concentrations (IC50s) observed suggest that MCF-10AP and MCF-10AH1047R cells with forced expression of mutant HRASQ61R were less sensitive to AKT inhibition than those cells expressing EV, whereas EV- and mutant HRASQ61R-expressing MCF-10AP and MCF-10AH1047R cells showed similar sensitivity to MEK inhibition (Supplementary Fig. 6c). In growth factor- and serum-free media, forced expression of mutant HRASQ61R in MCF-10AP and MCF-10AH1047R cells revealed a significant increase in cell proliferation (Fig. 7b, P < 0.0001, unpaired t-test), which was inhibited in part by either AKT or MEK inhibition regardless of the PIK3CA status. Importantly, however, the combination of both inhibitors completely abrogated cell proliferation (Fig. 7c).

29To assess the effects of AKT and MEK inhibition further, we treated EV- or mutant HRASQ61R-expressing MCF-10AP and MCF-10AH1047R cells grown in three-dimensional cultures. AKT and MEK inhibition led to a partial reversion of the phenotypic transformation caused by mutant HRASQ61R expression in MCF-10AP and MCF-10AH1047R cells, which was more overt under treatment with the combination of AKT and MEK inhibitors (Fig. 7d), consistent with the results obtained in monolayer cultures.

30Taken together, these findings demonstrate that HRASQ61R induces strong activation of the PI3K-AKT pathway in non-malignant breast epithelial cells; however, both PI3K-AKT and MAPK pathways likely contribute to RAS-mediated proliferation in these cells.

31Here we demonstrate that breast adenomyoepitheliomas constitute a heterogeneous group of tumors, characterized by recurrent pathogenic somatic mutations in HRAS and PI3K-AKT pathway genes. We further show that the histologic and genetic features of adenomyoepitheliomas vary according to ER status: ER-negative adenomyoepitheliomas more frequently display histologic features associated with an aggressive clinical behavior and frequently harbor concurrent mutations in HRAS Q61 and PIK3CA or PIK3R1 (56%), whereas ER-positive adenomyoepitheliomas are largely underpinned by PIK3CA or AKT1 mutations (59%). Members of the PI3K-AKT pathway, including PIK3CA, PIK3R1, and AKT1, are frequently affected by somatic genetic alterations in breast cancer and breast cancer precursor lesions. By contrast, HRAS Q61 hotspot mutations are vanishingly rare in breast cancer,, suggesting that in the breast, the co-occurrence of HRAS Q61 and PI3K pathway gene mutations may be associated with an adenomyoepithelial phenotype. In fact, the sole HRAS Q61 mutation found in the TCGA and ICGC breast cancer datasets was present in a high-grade triple-negative breast cancer, which was TP53 wild-type and harbored both an HRAS Q61L and a PIK3CA H1047R mutations. One could posit that this invasive breast cancer may have constituted a triple-negative invasive carcinoma arising from an adenomyoepithelioma.

32HRAS and PI3K-AKT pathway mutations identified by massively parallel sequencing were found to be clonal in the vast majority of cases and truncal in the three cases subjected to sequencing analysis of different components (i.e., primary adenomyoepithelioma, carcinoma and/or metastatic lesions), consistent with the notion that these mutations likely constitute founder genetic events in the development of adenomyoepitheliomas. Given that two cases harbored clonal HRAS mutations but subclonal PIK3CA or PIK3R1 mutations (Supplementary Fig. 2), one could posit that the HRAS hotspot mutations may precede the mutations affecting PI3K pathway genes in the development of ER-negative adenomyoepitheliomas. The acquisition of additional genetic changes, such as TERT promoter mutations and CDKN2A homozygous deletions, may play a role in tumor progression. In fact, an exploratory, hypothesis-generating analysis of TERT gene promoter mutations and CDKN2A homozygous deletions in the adenomyoepitheliomas subjected to massively parallel sequencing revealed a significant association with the presence of a carcinoma (Fig. 2, TERT gene promoter mutations, P = 0.0307, and CDKN2A homozygous deletions, P = 0.0086, Fisher’s exact tests). Further larger studies are warranted to test whether the presence of TERT gene promoter mutations and of loss of p16 protein expression (i.e., the protein product of CDKN2A) may predict the behavior of ER-negative adenomyoepitheliomas.

33In vitro analyses demonstrated that forced expression of mutant HRASQ61R alone or in the presence of mutant PIK3CAH1047R or PIK3CAE545K results in oncogenic properties and acquisition of a myoepithelial-like phenotype in non-malignant, TP53 wild-type, ER-negative breast epithelial cells. Albeit not necessary for the acquisition of a myoepithelial phenotype, given that adenomyoepitheliomas lacking HRAS mutations were observed in this study, forced expression of HRASQ61R was sufficient to induce a partial myoepithelial phenotype. Depending on the context and culture conditions, myoepithelial differentiation was more pronounced in cells harboring a knocked-in PIK3CAH1047R mutation than in PIK3CAWT cells, suggesting that these mutations may cooperate in the development and/or maintenance of the adenomyoepithelioma phenotype. In fact, our data suggest that mutant HRAS results in strong activation of the PI3K-AKT-mTOR signaling cascade, enhancing its signaling in a PIK3CA-mutant background (Fig. 7a, Supplementary Fig. 6a). Finally, given that the PIK3CA H1047R hotspot mutation, when introduced in the basal/myoepithelial compartment of mouse mammary glands, induces the development of tumors that often recapitulate adenomyoepitheliomas, further studies are required to define whether the likely cell of origin of most ER-negative breast adenomyoepitheliomas would reside in the basal compartment of the mammary gland (where progenitor cells likely reside), as opposed to the luminal progenitor compartment, the origin of the vast majority of breast cancers, including those of basal-like and triple-negative phenotype. Consistent with this notion, MMTV-Wnt1 mice have been shown to develop tumors that likely originate from progenitor cells and display several characteristics that resemble those of human ER-negative adenomyoepitheliomas, including a basal/myoepithelial transcriptomic profile, histologic features similar to those of adenomyoepitheliomas, and Hras Q61 mutations as a somatic event in up to 37% of cases, exclusively in Trp53 wild-type lesions. Conversely, the relevance of the MMTV-H-Ras mouse model for the study of adenomyoepitheliomas remains to be fully determined, given that these animals express wild-type rather than mutant Hras and have histologic features that appear to be distinct from those of human adenomyoepitheliomas, highlighting that caution should be exercised in the translation of the genotypic and phenotypic characteristics of mouse mammary gland tumors to human breast neoplasms.

34We and others have demonstrated that rare tumors that originate in distinct anatomical sites are not uncommonly underpinned by highly recurrent somatic genetic alterations (e.g., adenoid cystic carcinomas, which are driven by the MYB-NFIB or MYBL1 rearrangements in the breast, salivary glands and the lungs),. Adenomyoepitheliomas, in particular those displaying an ER-negative phenotype, bear a striking histologic similarity with epithelial−myoepithelial carcinomas of the salivary glands. Consistent with the notion that ER-negative adenomyoepitheliomas may constitute the breast counterpart of epithelial−myoepithelial carcinomas of the salivary glands, up to 80% of salivary gland epithelial−myoepithelial carcinomas have been reported to harbor HRAS Q61 hotspot mutations,, which are reported to co-occur with PIK3CA mutations in approximately 40% of cases. Taken together, the genomic analyses presented here and in previous studies,, and the in vitro data generated here suggest a potential genotypic−phenotypic association, where HRAS Q61 mutations in conjunction with PIK3CA or PIK3R1 mutations may result in the development of tumors with an epithelial−myoepithelial phenotype in certain anatomical sites.

35Our study has several limitations. Given that cases were obtained from multiple institutions and consultation files of the authors, we could not perform a systematic analysis of the impact of specific somatic genetic alterations on the outcome of patients with adenomyoepithelioma. A subset of adenomyoepitheliomas lacked mutations in HRAS and/or PI3K-AKT pathway genes, one of which harbored an ERBB3 hotspot mutation. Further studies are warranted to define the genomic drivers of adenomyoepitheliomas lacking HRAS and PI3K-AKT pathway gene mutations. Although we demonstrated here that the HRAS Q61R hotspot mutation, in particular in association with the PIK3CA hotspot mutations, may be sufficient for the acquisition of a myoepithelial phenotype in non-malignant breast epithelial cells and induces AKT signaling, the mechanistic basis for the acquisition of this myoepithelial differentiation program has yet to be defined. This would be ideally achieved using patient-derived adenomyoepithelioma cell lines, xenografts, and organoids, which are currently not commercially available. Finally, given that the PIK3CA H1047R hotspot mutation, when introduced in the basal/myoepithelial compartment of mouse mammary glands, induces the development of tumors that often recapitulate adenomyoepitheliomas, further studies are required to define whether the likely cell of origin of most ER-negative breast adenomyoepitheliomas would reside in the basal compartment of the mammary gland.

36In conclusion, breast adenomyoepitheliomas are heterogeneous and genomically distinct on the basis of their ER status. HRAS Q61 hotspot mutations and mutations affecting PI3K-AKT pathway-related genes likely constitute drivers of these tumors. The HRAS Q61R hotspot mutation was found to promote the acquisition of the cardinal features of ER-negative adenomyoepitheliomas in in vitro models, in particular in the presence of a PIK3CA H1047R or E545K hotspot mutation. Our findings contextualize the biological significance of HRAS Q61 hotspot mutations in the realm of breast neoplasms, and illustrate genotypic-phenotypic association in the taxonomy of breast tumors.

37After obtaining approval by the IRBs and the local research ethics committees from the authors’ institutions, representative histologic formalin-fixed paraffin-embedded (FFPE) blocks of 53 adenomyoepitheliomas of the breast were retrieved from the archives of the authors’ institutions. Patient consent was obtained where appropriate, according to the IRB-approved protocols. Samples were anonymized prior to the analyses. After central histologic review by five pathologists with expertise in breast pathology (F.C.G., M.E., I.O.E., E.A.R., and J.S.R.-F.), 43 cases were unanimously diagnosed as breast adenomyoepithelioma following the World Health Organization classification (WHO) criteria, and were included in this study (Supplementary Data 1). With an estimated mutation rate of 20 non-synonymous somatic mutations affecting protein coding genes per case and using Sidak correction (5% overall error rate), a sample size of 43 would confer 80% power for the detection of a recurrent genetic alteration if its true incidence is ≥15%. This study is compliant with the Declaration of Helsinki.

38Histologic assessment of architectural subtype, necrosis, mitotic rate, nuclear pleomorphism, and associated carcinomas were performed by three pathologists (F.C.G., F.P., and J.R.S.-F.). The architectural subtype was defined as tubular or papillary following previously defined criteria,. Necrosis, mitotic rate, and nuclear pleomorphism have been shown to constitute histologic features associated with aggressive behavior and malignant transformation in breast adenomyoepitheliomas,. Necrosis was defined as present (any area) or absent. Mitotic rate was defined as the number of mitotic figures in the myoepithelial or epithelial cell compartments per mm2, and stratified into three categories. Nuclear pleomorphism was evaluated according to the Nottingham histologic grading system of breast cancer. The presence and histologic type of an associated invasive carcinoma in the primary adenomyoepitheliomas and/or recurrent lesions in the ipsilateral breast was assessed according to the WHO criteria.

39Representative 4-μm-thick FFPE tumor sections of each case were subjected to immunohistochemical analysis with antibodies against ER (prediluted, clone 6F11, antigen retrieval ER1 solution for 20 min, Leica), and selected cases were immunohistochemically analyzed for the expression of E-cadherin (prediluted, clone 36, antigen retrieval CC1 for 16 min, Ventana), vimentin (prediluted, clone V9, antigen retrieval CC1 for 32 min, Ventana), HER2 (prediluted, 4B5, CC1 for 20 min, Ventana), p63 (prediluted, clone 4A4, antigen retrieval CC1 for 30 min, Ventana), p53 (prediluted, clone DO-7, antigen retrieval CC1 for 27 min, Ventana), Ki67 (1:400, clone MIB-1, antigen retrieval CC1 for 30 min, Dako), AR (1:500, clone SP107, antigen retrieval CC1 for 30 min, Ventana) and calponin (prediluted, clone EP798Y, antigen retrieval CC1 for 16 min, Ventana). Positive and negative controls were included in each slide run. The ER and HER2 status was defined according to the American Society of Clinical Oncology (ASCO)/College of American Pathologists (CAP) guidelines,. The expression of E-cadherin, vimentin, calponin, and p63 was analyzed as previously described. AR expression was defined as positive when ≥1% of tumor cells displayed nuclear expression; for statistical comparisons, we used a cut-off of ≥10% that has been previously employed to select patients for anti-androgen therapy in an early-phase clinical trial.

40Eight-µm-thick sections of FFPE blocks representative of the tumor and normal tissues (i.e. unaffected lymph nodes or breast tissue devoid of terminal duct-lobular units away from the tumor site) were stained with nuclear fast red and microdissected using a sterile needle under a stereomicroscope (Olympus SZ61), to ensure a tumor cell content >80% and that the normal tissue was devoid of any neoplastic cells as previously described. DNA extraction from microdissected tumor samples and normal adjacent tissues was performed separately using the DNeasy Blood and Tissue Kit (Qiagen), according to the manufacturer’s guidelines. We obtained DNA of sufficient quantity and quality of 10, 21, and 12 samples of primary tumors for WES, MSK-IMPACT and Sanger sequencing only, respectively (Supplementary Data 1). In addition, we obtained sufficient DNA from separately microdissected components of the primary tumor, local recurrence in the breast tissue, and/or metastases for three cases (AM5, AM8, and AM46). DNA quantity and quality were analyzed using a Qubit Fluorometer (Invitrogen, ThermoFisher) and a TapeStation (Agilent), respectively.

41WES and MSK-IMPACT, a massively parallel sequencing assay targeting all exons and selected non-coding and regulatory regions of 410 key cancer genes, were performed at the Memorial Sloan Kettering Cancer Center (MSKCC) Integrated Genomics Operation (IGO) on matched tumor and normal DNA samples from 10 and 21 adenomyoepitheliomas, respectively (Supplementary Data 1), as previously described,.

42Analyses of the sequencing data and the detection of somatic mutations and allele-specific CNAs were performed exactly as previously described,. In brief, reads were aligned to the human reference genome GRCh37 using the Burrows−Wheeler Aligner. Local realignment, de-duplication, and quality score recalibration were performed using the Genome Analysis Toolkit (GATK). Somatic single nucleotide variants (SNVs) were identified using MuTect; small insertions and deletions (indels) were identified using Strelka and VarScan 2,, and further curated by manual inspection. Variants found with >5% global minor allele frequency in dbSNP (Build 137) or that were covered by <10 reads in the tumor or <5 reads in the germline were disregarded,. Variants for which the tumor variant allele fraction was <5 times that of the normal variant allele fraction were disregarded,. We adopted this conservative approach to minimize false positive results obtained with DNA extracted from FFPE samples. The potential functional effect of each missense SNV was investigated following our previously described approach,, using a combination of benchmarked mutation effect algorithms. Hotspot mutations were annotated according to Chang et al.. Allele-specific CNAs were inferred from WES or MSK-IMPACT data using FACETS as previously described,. In brief, read counts for dbSNP (build 137) positions within the target regions with dbSNP entries (build 137) were generated for matched tumor and normal samples, and used as input to FACETS, which performs a joint segmentation of the total and allelic copy ratio and infers allele-specific copy number states, using the following parameters: pre-processing critical value (Pre CVAL) 50, critical value for estimating diploid Log2 ratio (CVAL1) 150, critical value for segmentation (CVAL2) 50, and minimum number of heterozygous SNPs in a segment used for bivariate t-statistic during clustering of segment (Min Nhet) 25. Genes with CNAs were determined adopting the methods described in Curtis et al. and in the supplementary materials of Piscuoglio et al.. The cancer cell fraction (CCF) of each mutation was inferred using ABSOLUTE (v1.0.6), as previously described,,. Solutions from ABSOLUTE were manually reviewed as described,. A mutation was classified as clonal if its probability of being clonal was >50% or if the lower bound of the 95% confidence interval of its CCF was >90%. Mutations that did not meet the above criteria were considered subclonal. Phylogenetic trees were generated using Treeomics, using the CCF values and the depth as input.

43A random subset of 66 somatic mutations, encompassing 78% of the non-synonymous mutations identified by WES in samples AM1−AM7 (excluding case AM5 and its associated lesions, see below) were validated by targeted amplicon re-sequencing using a custom Ion Torrent AmpliSeq panel. Sequencing was performed to a median depth of 832× (range 560×−941×) and 664× (range 257×−949×) for the tumor and germline samples, respectively. Paired-end reads in FASTQ format were aligned to the reference human genome GRCh37 using the Torrent Mapping Alignment Program (TMAP, v3.4.1). Local realignment was performed using GATK (v3.1.1). Putative mutations were interrogated using pileup files generated with samtools mpileup (version 1.2 htslib 1.2.1). Mutations present at variant allele frequencies >1% were considered “validated”. Mutations that did not validate were excluded from further analyses. Given the overall validation rate of 92.4% (Supplementary Data 2), mutations that were not tested were included in the final results. For AM5 and its associated lesions, validation was performed through independent WES assays from new libraries generated from the original DNA samples of each lesion; only mutations found in both WES assays from each component were considered as validated.

44For validation of the HRAS, PIK3CA, and AKT1 mutations detected in the samples subjected to WES or MSK-IMPACT sequencing and for the screening of mutations affecting these genes in additional 12 samples not subjected to massively parallel sequencing, PCR amplification and Sanger sequencing were performed as previously described, (primer sequences upon request). In addition, the TERT gene promoter was assessed either by MSK-IMPACT or Sanger sequencing, using the primers and methods described in Piscuoglio et al.. Sequencing reactions were performed in triplicate, and both the forward and reverse strands were analyzed using MacVector software.

45For western blotting, we employed antibodies against HRAS (Santa Cruz Biotechnology sc520, rabbit, 1:500), αSMA (Abcam ab5694, rabbit, 1:500), CK5 (ThermoFisher MA5-17057, mouse, 1:500), and p63 (Biolegend 619002, rabbit, 1:500), and the following antibodies purchased from Cell Signaling Technology (CST): rabbit anti-p-MEK (S217/221; 9154, 1:1000), anti-p-ERK1/2 (T202/Y204; 4370, 1:1000), anti-ERK1/2 (9107, 1:1000), anti-p-p90 RSK (S380; 9341, 1:1000), anti-p-AKT (S473; 4060, 1:1000), anti-p-AKT (T308; 2965, 1:1000), anti-AKT (9272, 1:1000), anti-p-PRAS40 (T246; 13175, 1:1000), anti-p-FOXO1/3a/4 (2599, 1:1000), anti-p-GSK3β (S9; 5558, 1:1000), anti-p-mTOR (S2448;5536, 1:1000), anti-p-p70 S6K (T389; 9205, 1:1000), anti-p-S6 (S240/244; 5364, 1:1000), anti-p-4EBP1 (S65; 9451, 1:1000), anti-p-4EBP1 (T37/46; 2855, 1:1000), anti-TIMP1 (8946, 1:1000), anti-vimentin (5741, 1:1000), anti-α-Tubulin (2125, 1:1000), anti-β-actin (4970, 1:1000) and anti-Histone H3 (4499, 1:1000), and mouse anti-E-cadherin (14472, 1:1000) and anti-α-Tubulin (3873, 1:1000). For immunofluorescence, antibodies against vimentin (CST 5741, rabbit, 1:100), E-cadherin (CST 14472, mouse, 1:50), and calponin (Abcam ab46794 rabbit, 1:100) were used.

46The AKT inhibitor MK2206 (MedChem Express HY-10358) and MEK inhibitor GSK212 (Selleck Chemicals s2658) were employed for pharmacological inhibition in cell lines analyzed in in vitro experiments time points and concentrations indicated.

47The human HRAS (NM_005343) cDNA ORF clone was purchased from Origene (RG216409), and the Q61R mutation was introduced using the Q5 Site-Directed Mutagenesis Kit (New England Biolabs E0554) following the manufacturer’s recommendations. HRASWT and mutant HRASQ61R open reading frames were cloned into the pLenti-EF1a-GFP-2A-Puro vector (ABM LV067) and pcDNA™3.3-TOPO® vector (ThermoFisher K830001) for stable and transient transfection, respectively. Sanger sequencing was used to confirm the reading frames of HRASWT and HRASQ61R. Human isogenic MCF-10A PIK3CA wild-type (MCF-10AP) and MCF-12A cells were purchased from ATCC. PIK3CA H1047R mutant (MCF-10AH1047R) and E545K mutant (MCF-10AE545K) cells were purchased from Horizon (X-MAN). Cell lines were authenticated by short tandem repeat profiling as previously described, and tested for mycoplasma infection using the Universal Mycoplasma Detection Kit (ATCC). Cell lines were cultured as previously described. Transfections of EV, HRASWT, and mutant HRASQ61R were performed as previously described,, using Lipofectamine™ LTX Reagent (ThermoFisher 15338100) according to the manufacturer’s protocol. For stable selection, transfected cells were selected with 1 µg/ml puromycin. Resistant colonies formed at 15−20 days of selection.

48Pull-down of the active form of HRAS from stable EV, HRASWT, and mutant HRASQ61R MCF-10AP, MCF-10AH1047R, and MCF-12A cell lysates was performed using two independent active HRAS pull-down and detection kits from Cell Biolabs, Inc. (STA-400-H-T) and ThermoFisher (16117), respectively. Briefly, 80–90% confluent cells from each group were scraped and collected using the lysis/wash buffer provided. 40 µl to 80 µl of resuspended GST-Raf1-RBD Agarose beads were mixed into each cell lysate (containing at least 500 µg of total protein) and incubated at 4 °C for 1 h followed by three washes with 0.5 ml of lysis/wash buffer. Subsequently, the bead pellet was resuspended in 40−50 µl of 2× reducing SDS-PAGE sample buffer (β-mercaptoethanol-2× SDS sample buffer 1:20). After centrifugation, the eluted samples were boiled for 5 min and loaded onto a polyacrylamide gel for GTP-HRAS detection by western blot analysis using an anti-HRAS antibody.

49Total protein and nuclear protein lysates were prepared using the M-PER Mammalian Protein Extraction Reagent and NE-PER Nuclear and Cytoplasmic Extraction Reagents, respectively, supplemented with Halt Protease and Phosphatase inhibitors cocktail (ThermoFisher). Standard western blotting was conducted as previously described. Membranes were probed with primary antibodies and followed by incubation with HRP-tagged (CST 7074) or conjugated IRDye 680RD/800CW (LI-COR Biosciences) secondary antibodies, visualized on a Syngene ChemiGenius with Super-Signal West Dura Chemiluminescence Substrate (Pierce) or the Odyssey Infrared Imaging System, and quantified by the LI-COR Image Studio Software. When required, the membranes were stripped using the Restore PLUS Western Blot Stripping Buffer (ThermoFisher). Replicate experiments were performed as indicated. Unprocessed scans of blots are available in Supplementary Figs. 7−10.

50Soft agar colony formation assay was performed as previously described,. Briefly, MCF-10AP, MCF-10AH1047R, and MCF-12A stably expressing EV, HRASWT, and mutant HRASQ61R (5×104) were added to 1.5 ml of complete growth media with 0.4% UltraPure™ Agarose (ThermoFisher) and layered onto a 2 ml bed of complete growth media plus 0.5% of agarose. Cells were fed every 3 days with 1 ml of complete growth media. At day 21, growth media was removed and viable colonies were stained with 0.005% Crystal Violet solution (Sigma-Aldrich). Colony size was determined using Fiji (ImageJ). Size (in pixels) was measured using Feret diameter (D) and minimum Feret diameter (d) and applying the formula (D x d2)/2 = colony volume and plotted with GraphPad Prism v_7.0a. Experiments were performed in triplicate.

51For cell proliferation assays, cells were seeded either in completed growth media or growth factor- and serum-free media in 96-well plates, and monitored over 4 or 7 days, as indicated. Experiments were performed in triplicate. For AKT and MEK pharmacological inhibition, cells were seeded in complete media or growth factor- and serum-free media, and inhibitors were added the following day. 25 µl of Resazurin (R&DSystems, AR002) was added to each well and incubated at 37 °C for 4 h. Absorbance was read at spectra of 560EX nm/590EM nm using SpectraMax M5 (Molecular Devices). Growth and inhibition curves were plotted and analyzed using GraphPad Prism v_7.0a. To determine the median inhibitory concentration (IC50) of the AKT inhibitor and MEK inhibitor, mean values of the number of cells treated with the indicated inhibitors for 4 days were plotted as percentage of inhibition against the Log concentration of inhibitors (nM), and nonlinear regression analysis was performed using GraphPad Prism v_7.0a. Triplicate experiments were repeated at least twice.

52Cells were serum starved for 16 h in 2% horse serum DMEM/F12 media without EGF, and were trypsinized, seeded (15×105) and cultured overnight in six-well plates. A pipette tip was used to generate a scratch in the cell layer. Images were obtained at 0 and 24 h after scratch wounding at the same position. The percentage of wound closure indicated by scratch width reduction was assessed and plotted using GraphPad Prism v_7.0a. Experiments were performed in triplicate.

53Total RNA was extracted from cells grown in monolayer or three-dimensional cultures using TRIzol reagent (ThermoFisher) according to the manufacturer’s instructions. Two micrograms of the extracted total RNA from each sample were employed for cDNA synthesis using SuperScript VILO Master Mix (ThermoFisher). cDNA was amplified using the StepOnePlus Real-Time PCR System (ThermoFisher). Quantitative TaqMan RT-PCR (ThermoFisher) was performed for CDH1 (Hs01023894_m1), VIM (Hs00958111_m1), SPARC (Hs00234160_m1), TIMP1 (Hs00171558_m1), and ACTA2 (Hs00426835_g1). Experiments were performed in triplicate, and expression data were normalized to the housekeeping gene GAPDH (Hs99999905_m1) and calculated as 2–[(Ct of gene) − (Ct of GAPDH)].

54Cells were seeded on top of growth factor-reduced reconstituted basement membrane (Matrigel, BD Biosciences) as previously described,, and subjected to immunofluorescence and mRNA expression analysis on day 10. Briefly, 50 µl or 200 µl of Matrigel was added to coat each well of an eight-well chamber slide (for IF staining and pharmacological inhibition) or a 24-well tissue culture plate (for RNA extraction), respectively. Five thousand cells resuspended in 500 µl of assay media (with 5 ng/ml EGF and 2% Matrigel) or 20,000 cells resuspended in 1 ml assay media were plated on top. For pharmacological inhibition, cells were treated with DMSO or the indicated inhibitors 3 days after seeding in drug-free media. Fresh media with/without DMSO or the indicated inhibitors was replenished every 3 days. Experiments were repeated at least twice.

55Immunofluorescence analysis of three-dimensional organotypic cultures was performed as previously described. Acinar structures in eight-well chambers were fixed and permeabilized in 4% PFA and 0.5% TritonX-100, respectively. After 1 h blocking in 10% goat serum, primary antibody was added for overnight incubation at 4 °C, and Alexa Fluor-conjugated secondary antibody (1:500) was added for 1 h incubation at room temperature. The slides were mounted in ProLong Gold antifade reagent with DAPI (ThermoFisher). Confocal analyses were performed with the Leica SP5 DM confocal microscopy system equipped with four lasers: an ultraviolet (UV) diode (405 nm), an argon laser (458, 476, 488, and 514 nm), a 543 nm HeNe laser, and a 633 nm HeNe laser. Experiments were performed in triplicate.

56Tumor/mammosphere assays were performed as previously described. In brief, cells were plated (30,000/well) as single cell suspensions in ultralow attachment six-well plates and grown in DMEM:F12 media (serum-free) supplemented with 20 µl/ml B27 (ThermoFisher), 20 ng/ml EGF, and 20 ng/ml bFGF. Fresh media (1 ml) was added every 3 days. Mammospheres were counted and photographed at day 10. Experiments were performed in triplicate. Colony and sphere images were documented using the phase contrast EVOS XL Imaging System (ThermoFisher). Sphere number was determined using Fiji (ImageJ).

57Fisher’s exact tests, Wilcoxon tests, and unpaired t-tests were used for the comparison of categorical, non-parametric, and continuous parametric variables, respectively. For the comparisons of continuous data, we assessed whether the variables were heteroscedastic and utilized appropriate statistical methods accordingly. Mutual exclusivity between PIK3CA, PIK3R1, and AKT1 was assessed using CoMEt exact test. Co-occurrence between HRAS mutations and PIK3CA or PIK3R1 mutations was assessed using a Z test. Statistical analyses were carried out using R v3.1.2 or GraphPad Prism v_7.0a. Two-tailed P values <0.05 were considered statistically significant. For all experiments, 95% confidence intervals were adopted.

58WES and MSK-IMPACT sequencing data have been deposited in the NCBI Sequence Read Archive under the accession numbers SRP065277 and SRP065302, respectively. The publicly available dataset from the TCGA breast cancer study was retrieved from the cBioPortal website (www.cBioPortal.org) and TCGA data portal (https://tcga-data.nci.nih.gov/docs/publications/tcga/) on 07/03/2017. Somatic mutations from samples included in the ICGC breast cancer study were extracted from the supplementary materials of Nik-Zainal et al.. All other remaining data are available within the article and Supplementary Files, or available from the authors upon request.